Re: [PATCH] btrfs-progs: convert, warn if converting a fs which won't mount

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2020/3/4 上午10:14, Anand Jain wrote:
>
>
> On 3/4/20 1:44 AM, David Sterba wrote:
>> On Fri, Feb 28, 2020 at 05:06:52PM +0800, Anand Jain wrote:
>>> On 2/28/20 4:27 PM, Qu Wenruo wrote:
>>>> On 2020/2/28 下午4:03, Anand Jain wrote:
>>>>> On aarch64 with pagesize 64k, btrfs-convert of ext4 is successful,
>>>>> but it won't mount because we don't yet support subpage blocksize/
>>>>> sectorsize.
>>>>>
>>>>>    BTRFS error (device vda): sectorsize 4096 not supported yet,
>>>>> only support 65536
>>>>>
>>>>> So in this case during convert provide a warning and a 10s delay to
>>>>> terminate the command.
>>>>
>>>> This is no different than calling mkfs.btrfs -s 64k on x86 system.
>>>> And I see no warning from mkfs.btrfs.
>>>>
>>>> Thus I don't see the point of only introducing such warning to
>>>> btrfs-convert.
>>>>
>>>
>>> I have equal weight-age on the choices if blocksize != pagesize viz..
>>>     delay and warn (this patch)
>>>     quit (Nikolay).
>>>     keep it as it is without warning (Qu).
>>>
>>>    Here we are dealing with already user data. Should it be different
>>>    from mkfs?
>>>    Quit is fine, but convert tool should it be system neutral?
>>
>> The delays should be used in exceptional cases, now we have it for check
>> --repair and for unfiltered balance. Both on user request because
>> expecting users to know everything in advance what the commands do has
>> shown to be too optimistic.
>>
>> Refusing to allow the conversion does not make much sense for usability,
>> mising the unmounted and mounted constraints.
>>
>> A warning might be in place but there's nothing wrong to let the user do
>> the conversion.
>>
>> I've tried mkfs.ext4 with 64k block size and it warns and in the
>> interactive session wants to confirm that by the user:
>>
>>    $ mkfs.ext4 -b 64k img
>>    Warning: blocksize 65536 not usable on most systems.
>>    mke2fs 1.45.5 (07-Jan-2020)
>>    img contains a ext4 file system
>>       created on Tue Mar  3 18:41:46 2020
>>    Proceed anyway? (y,N) y
>>    mkfs.ext4: 65536-byte blocks too big for system (max 4096)
>>    Proceed anyway? (y,N) y
>>    Warning: 65536-byte blocks too big for system (max 4096), forced to
>> continue
>>    Creating filesystem with 32768 64k blocks and 32768 inodes
>>
>>    Allocating group tables: done
>>    Writing inode tables: done
>>    Creating journal (4096 blocks): done
>>    Writing superblocks and filesystem accounting information: done
>>
>
> Just warn is reasonable. But I don't think you meant to introduce
> interactive part similar to mkfs.ext4 in btrfs-convert? we don't have it
> anywhere in btrfs-progs. As the btrfs-convert is not an exceptional case
> (though it deals with the user data) removing the delay makes sense,
> mover over the conversion and the rollback does not take much time in
> general.
>
> Thanks, Anand

+1 for warning only, especially when btrfs-convert is revertable, unlike
mkfs.

Thanks,
Qu